Martinsos
commented
10 years ago Hi @rob-p , thank you for your interest! They are somewhat overlapping in functionality and have similar interfaces, but also have some big differences and are useful for different purposes. Main differences:
- Swimd is 1 vs many (database search) aligner, while Edlib is 1 vs 1 aligner. That means that Edlib can be used to compare two sequences to each other, while Swimd should be used to compare one sequences against many sequences, otherwise there will be no speedup. Swimd does comparison with many sequences at the same time, so for example if it can fit 16 sequences in its register at the same time, comparing with less than 16 sequences will not give as big speedup as it could achieve. So more the sequences to compare with, better.
- Swimd uses SSE/AVX2 instructions of CPU, while Edlib does not. Most of CPUs today have SSE instructions so this is maybe not so important.
- Swimd can do local alignment, global alignment or semi-global alignment: SW, NW, HW or OV, while Edlib can do global or semi-global alignment: NW, HW or SHW. Main difference here is that Edlib can not do local alignment. Each of this alignment methods is most suitable for certain situation.
- Maybe the most important difference: Edlib uses edit distance (Levenstein distance) for scoring and can not use any other scoring system, while Swimd's scoring is defined by scoring matrix and gap penalties which are specified as input. That makes Swimd much more flexible, and that is also the reason why Edlib does not offer local alignment or OV semi-global method: local alignment and OV do not go with edit distance. On the other hand, often edit distance is good enough, so in that case it makes more sense to use Edlib, as it is inherently much faster than more flexible aligners like Swimd.
- For now, Edlib offers finding of alignment path, while Swimd does not. I plan to implement finding of alignment path for Swimd also, so I hope this will change soon.
To summarize: if edit distance is enough, Edlib is probably better choice as it is much faster, otherwise use Swimd. In case edit distance is not enough and you also need pairwise alignment (1 vs 1, which Swimd is not good for) and not database search, I recommend using @mengyao 's library: https://github.com/mengyao/Complete-Striped-Smith-Waterman-Library . It is slower than Swimd for database search, but can be used for alignment 1 vs 1. I hope I answered your question, if there is anything else or if I was not clear enough, please ask!
rob-p
Hi Martinos,
I'm interested in potentially making use of one of your libraries, and, in my searching, I've come across the repositories for both swimd and edlib. These seem like they have very similar (or at least overlapping) functionality. Could you elaborate a little bit on what the differences are and which library you'd recommend for different use cases?
Thanks! Rob